Re: [RFC PATCH 1/3] sched: add sched_policy and it's sysfs interface

[Alex Shi <alex.shi@intel.com>]

On 11/07/2012 10:41 PM, Greg KH wrote:
> On Wed, Nov 07, 2012 at 08:27:17PM +0800, Alex Shi wrote:
>>>> diff --git a/drivers/base/cpu.c b/drivers/base/cpu.c
>>>> index 6345294..5f6a573 100644
>>>> --- a/drivers/base/cpu.c
>>>> +++ b/drivers/base/cpu.c
>>>> @@ -330,4 +330,6 @@ void __init cpu_dev_init(void)
>>>>  		panic("Failed to register CPU subsystem");
>>>>  
>>>>  	cpu_dev_register_generic();
>>>> +
>>>> +	create_sysfs_sched_policy_group(cpu_subsys.dev_root);
>>>
>>> Are you sure you didn't just race with userspace, creating the sysfs
>>> files after the device was created and announced to userspace?
>>
>> Sorry for don't fully get you. Is the sysfs announced to userspace
>> just in 'mount -t sysfs sysfs /sys'?
> 
> No, when the struct device is registered with the driver core.
> 
>> The old powersaving interface: sched_smt_power_savings also
>> created here. and cpu_dev_init was called early before do_initcalls
>> which cpuidle/cpufreq sysfs were initialized.
>>
>> Do you mean this line need to init as core_initcall?
> 
> No, you need to make this as an attribute group for the device, so the
> driver core will create it automatically before it tells userspace that
> the device is now present.
> 
> Use the default attribute groups and you should be fine.

Thanks a lot for explanation! :)

It seems a misunderstanding here. I just create a sysfs group, no device
registered.
The code followed the cpuidle's implementation:

$ ls /sys/devices/system/cpu/cpuidle/
current_driver  current_governor_ro

Seems it's still better to move the group creation into sched/fair.c not
here.
> 
> Hope this helps,
> 
> greg k-h
> 


-- 
Thanks
    Alex